A control method for a dividing wall column for high purity silane gas production

Abstract

The application relates to the technical field of rectifying column control, in particular to a method for controlling a partition rectifying column for preparing high-purity silane gas, which comprises the following steps: collecting various parameter data of the partition rectifying column in real time during the preparation of the high-purity silane gas; obtaining a comprehensive feature matrix of the partition rectifying column by the abnormality degree of the various parameter data at each collection time and the difference between the abnormality degrees of different parameter data; dividing the whole time period of the collected parameter data into time intervals; obtaining the characteristic values and characteristic coefficients of the various parameter data in the time intervals; and obtaining the control coefficients of the various parameter data and controlling the various parameters of the partition rectifying column. The application aims to improve the timeliness of controlling the various parameters of the partition rectifying column by adjusting the value of the proportional parameter of the PID controller when controlling the various parameters, thereby improving the stability of the various parameters and the preparation quality of the high-purity silane gas.

Description

Technical Field

[0001] This application relates to the field of distillation column control technology, specifically to a control method for a partition distillation column used in the preparation of high-purity silane gas. Background Technology

[0002] With the rapid development of industries such as semiconductors and photovoltaics, the market demand for silane gas, as a key electronic gas, continues to rise. In semiconductor chip manufacturing, silane is used in chemical vapor deposition (CVD) processes to grow high-quality silicon thin films, and its purity directly affects the performance, size, and integration density of chip transistors. In the photovoltaic field, high-purity silane gas can improve the photoelectric conversion efficiency and lifespan of solar cells. The partition distillation column, with its unique internal structural design, significantly reduces backmixing between different components within the column. Compared to traditional distillation columns, it improves separation efficiency, achieves more precise material separation, and reduces energy consumption, gradually becoming a key piece of equipment in the preparation of high-purity silane gas.

[0003] Currently, PID controllers are commonly used to stably control parameters such as temperature, pressure, and flow rate in diaphragm distillation columns, maintaining their normal operation. However, in the preparation of high-purity silane gas, there is a strong coupling relationship between the parameters of the diaphragm distillation column. If one parameter deviates from the set value due to disturbance, the other parameters may exhibit complex dynamic responses due to the coupling relationship. Traditional PID controllers struggle to quickly and accurately adapt to these complex parameter changes and cannot adjust the control strategy in a timely and effective manner, resulting in insufficient stability of the diaphragm distillation column parameters and affecting the quality of the prepared high-purity silane gas. Summary of the Invention

[0004] In view of the above, it is necessary to provide a control method for a diaphragm distillation column for the preparation of high-purity silane gas. Compared with the traditional control method for diaphragm distillation columns for the preparation of high-purity silane gas, this method improves the timeliness of the control of various parameters of the diaphragm distillation column by adjusting the value of the proportional parameter when the PID controller controls various parameters, thereby improving the stability of various parameters and thus improving the preparation quality of high-purity silane gas.

[0005] The method for controlling a baffled distillation column for the preparation of high-purity silane gas in this application adopts the following technical solution: One embodiment of this application provides a method for controlling a partitioned distillation column for the preparation of high-purity silane gas, the method comprising the following steps: Real-time acquisition of various parameter data of the baffle distillation column during the preparation of high-purity silane gas; By evaluating the degree of anomalies in various parameter data at each acquisition time, a parameter feature matrix of the baffled distillation column is constructed; by obtaining the difference weights of data in different rows of the parameter feature matrix, the difference weights of each row of data are obtained; and by combining the difference weights with the parameter feature matrix, a comprehensive feature matrix of the baffled distillation column is obtained. By analyzing the abrupt changes in the elements of the comprehensive feature matrix, the entire time period for collecting parameter data is divided into various time intervals. The feature values ​​of each parameter data within any given time interval are compared to the differences in the degree of abnormality of each parameter data within that time interval, thus obtaining the feature values ​​of each parameter data within that time interval. Furthermore, the feature factors of each time interval are obtained based on the degree of overlap between that time interval and the time period. Finally, the feature coefficients of each parameter data within that time interval are obtained by combining these feature values. By analyzing the distribution and characteristic coefficients of the anomalies of various parameter data within each time interval, control coefficients for each parameter data are obtained. The control coefficients, combined with a PID controller, are used to control various parameters of the baffle distillation column.

[0006] In one embodiment, the process of constructing the parameter feature matrix is ​​as follows: An anomaly detection algorithm is used to obtain the anomaly scores of each parameter data at each acquisition time. The anomaly scores of each parameter data at all acquisition times are arranged in chronological order to form parameter vectors for each parameter data, where the parameter vectors are row vectors. The parameter vectors of all parameter data are then combined to form a parameter vector matrix.

[0007] In one embodiment, the difference weight is a normalized value of the mean DTW distance between each row of data and all other rows of data in the parameter vector, wherein the sum of the difference weights of all rows of data in the parameter feature matrix is ​​1.

[0008] In one embodiment, the process of obtaining the comprehensive feature matrix is ​​as follows: The difference weights of all rows of data in the parameter feature matrix are arranged in row order to form a weight matrix. The comprehensive feature matrix is ​​the product of the weight matrix and the parameter feature matrix, where the weight matrix is ​​a row vector.

[0009] In one embodiment, the process of dividing the time interval is as follows: Arrange all elements in the comprehensive feature matrix in chronological order to form a feature sequence; Each mutation point in the feature sequence is obtained, and the time period is divided into time intervals based on the time of the mutation point.

[0010] In one embodiment, the process of obtaining the feature value is as follows: Arrange all values ​​of the comprehensive feature matrix within any time interval in chronological order to form a first vector; Arrange all the abnormal scores of each parameter data within any time interval according to time sequence to form a second vector of each parameter data; The DTW distance between the first vector and the second vector is denoted as the metric distance; The proportion of the metric distance corresponding to each parameter data in any given time interval among all the metric distances is used as the feature value of each parameter data in that given time interval.

[0011] In one embodiment, the feature factor is the intersection-union ratio of the any time interval to the time period.

[0012] In one embodiment, the feature coefficient is the mean of the feature value and the feature factor.

[0013] In one embodiment, the process of determining the control coefficient is as follows: Calculate the average of all the aforementioned anomaly scores for each parameter data within each time interval; Calculate the product of the characteristic coefficients of each parameter data in each time interval and the average value; The control coefficient is the sum of the products of the various parameter data over all time intervals.

[0014] In one embodiment, the parameters of the control diaphragm distillation column include: The initial value of the proportional parameter of the PID controller when controlling various parameters is obtained by the PID parameter tuning method. The sum of the initial value and the normalized value of the control coefficient of each parameter data is used as the optimized value of the proportional parameter of the PID controller when controlling various parameters. The optimized values ​​are used to control various parameters of the baffle distillation column.

[0015] This application has at least the following beneficial effects: This application analyzes the degree of anomaly of various parameters at each acquisition time and constructs a parameter feature matrix to centrally reflect the anomaly characteristics of the parameter data. By calculating the difference weights of each row of data in the parameter feature matrix, the differences between different parameter data are quantified, providing a weight basis for subsequent comprehensive analysis. By combining the difference weights with the parameter feature matrix, a comprehensive feature matrix is ​​obtained, which can comprehensively reflect the comprehensive characteristics of the anomaly changes of the baffle distillation column at different acquisition times. Through the application of difference weights, the reasonable weights of various parameter data in the comprehensive analysis are ensured, improving the accuracy of the analysis of the comprehensive characteristics of the anomaly changes at each acquisition time. Furthermore, by comprehensively analyzing the abrupt changes of elements in the feature matrix, the time intervals of abnormal changes caused by interference and coupling relationships are precisely segmented. By comparing the abnormal changes of various parameter data within the segmented time intervals with the overall characteristics, the control coefficients of various parameter data of the baffle distillation column are determined. The control coefficients can reflect the interference influence and coupling characteristics of various parameter data in the actual control process, providing a basis for optimizing the control of the baffle distillation column. Through the control coefficients, the proportional parameters of the PID controller are adjusted. By accurately analyzing the characteristics of untimely actual control response caused by strong coupling relationships under the influence of interference, the timeliness of controlling various parameters of the baffle distillation column can be improved, the stability of various parameters of the baffle distillation column can be improved, and thus the quality of high-purity silane gas preparation can be improved. Attached Figure Description

[0016] Figure 1 A flowchart illustrating the steps of a partition distillation column control method for the preparation of high-purity silane gas provided in this application; Figure 2 This is a schematic diagram of a partition distillation column; Figure 3 A schematic diagram of the time interval segmentation process; Figure 4 This is a schematic diagram illustrating the process of obtaining control coefficients.

[0017] Figure 2 1 is the stripping section, 2 is the rectification section, 3 is the pre-separation section, 4 is the side stream extraction section, 5 is the feed inlet, 6 is the top extraction outlet, 7 is the side stream extraction outlet, and 8 is the bottom extraction outlet. Detailed Implementation

[0018] The following description, in conjunction with the accompanying drawings, details a specific scheme for controlling a partition distillation column used in the preparation of high-purity silane gas, as provided in this application.

[0019] This application provides a method for controlling a partitioned distillation column for the preparation of high-purity silane gas, specifically, the following method is provided. Please refer to [link to relevant documentation]. Figure 1 The method includes the following steps: Step 1: Real-time acquisition of various parameter data of the baffle distillation column during the preparation of high-purity silane gas.

[0020] The feed gas containing silane is passed through a filter to remove solid particulate impurities. Then, a molecular sieve is used as an adsorbent to dry the feed gas to remove moisture and prevent moisture from reacting with silane in the subsequent distillation process.

[0021] In this embodiment, the raw material gas is obtained through a preliminary reaction using the magnesium silicide method. In addition to silane, the raw material gas also contains impurities such as hydrogen, nitrogen, and moisture.

[0022] A schematic diagram of the structure of a baffle distillation column is shown below. Figure 2 As shown, the filtered and dried feed gas enters the pre-separation section 3 from the feed inlet 5 of the baffled distillation column. In the pre-separation section 3, the feed gas is initially separated into light and heavy components. The light components are mainly hydrogen and nitrogen, while the heavy components are mainly high-boiling-point impurities. In the rectification section 2, the rising gas phase undergoes sufficient mass and heat transfer with the descending liquid phase. During this process, the light components, mainly silanes, in the gas phase are continuously enriched. Through multiple contacts with the trays or packing, the concentration of silanes gradually increases. Simultaneously, the descending liquid phase enters the stripping section 1, where the heavy components are further concentrated and finally separated from the bottom of the column. High-purity silane products are separated through the side stream outlet 7; the light component gas is discharged at the top outlet 6, which can be recycled or treated for emission as needed; the heavy component liquid is discharged at the bottom outlet 8, which is periodically treated for emission.

[0023] Thermocouple temperature sensors are installed at the top, bottom, feed inlet 5, side outlet 7, and both sides of the partition in the partition distillation column to measure the temperature in real time and monitor the heat distribution and mass transfer within the column. Pressure sensors are used to measure the pressure data at the top and bottom of the column in real time to monitor the pressure within the column and prevent pressure fluctuations from affecting the distillation effect. Mass flow meters are installed on the feed gas inlet pipe, top outlet pipe, bottom outlet pipe, side outlet pipe, and reflux pipe to measure the flow rate data of various materials in real time. It should be noted that temperature, pressure, and flow rate are all different parameters at different locations.

[0024] In this embodiment, the thermocouple temperature sensor, pressure sensor and mass flow meter all collect data once per minute. The value of the collection frequency is preset by the user and can be set by the implementer. This application does not impose any special restrictions.

[0025] Step 2: By analyzing the variation characteristics of various parameter data of the partition distillation column during the preparation of high-purity silane gas, the control coefficients of various parameter data of the partition distillation column are obtained.

[0026] Because the conditions inside the partition distillation column are complex during the actual preparation process, the collected data may be affected by interference, resulting in low data quality. Therefore, this application uses a filter to reduce noise in the collected parameter data of the partition distillation column, thereby reducing the impact of interference factors on the parameter data and improving the accuracy of subsequent analysis of the control response of the partition distillation column parameters.

[0027] In this embodiment, a Wiener filter is used to reduce noise in the parameter data. As another implementation method, other existing feasible filters can be used to reduce noise in the parameter data.

[0028] Under normal circumstances, during the preparation of high-purity silane gas, the parameters of various parts of the baffled distillation column remain stable when undisturbed. However, when subjected to disturbances, if the PID controller's response to a certain parameter is not timely, the strong coupling relationship between parameters within the baffled distillation column may cause significant fluctuations in other parameters, increasing the error in parameter control and prolonging the control response time of various parameters within the baffled distillation column. To address these issues, this application analyzes the variation characteristics of various parameter data in the baffled distillation column during the preparation of high-purity silane gas, extracting the mutual influence relationships between parameters during actual control to achieve precise optimized control adjustments.

[0029] Step 2.1: Construct a parameter feature matrix for the baffled distillation column by evaluating the degree of anomaly of various parameter data at each acquisition time; obtain the difference weight of each row of data by the differences between different rows of data in the parameter feature matrix; combine the difference weight with the parameter feature matrix to obtain the comprehensive feature matrix of the baffled distillation column.

[0030] In the preparation of high-purity silane gas, under normal operating conditions of the baffled distillation column, the various parameters typically do not exhibit significant correlation changes. However, when subjected to disturbances that cause significant and prolonged changes in certain parameters, within the timeframe of the disturbance, coupling relationships may exist between the parameters. A significant change in one parameter can trigger a chain reaction in other parameters, resulting in strong correlations within that timeframe. Based on these characteristics, this application aims to analyze the correlation changes of different parameters under disturbances caused by coupling relationships. An anomaly detection algorithm is used to obtain the anomaly scores of each parameter data at each acquisition time. The anomaly scores of each parameter data at all acquisition times are arranged chronologically to form parameter vectors for each parameter data, where each parameter vector is a row vector. The parameter vectors of all parameter data are then combined to form the parameter vector matrix of the baffled distillation column, collectively reflecting the anomaly changes of the parameter data over time.

[0031] In this embodiment, the LOF (Local Outlier Factor) anomaly detection algorithm is used to obtain the anomaly scores of each parameter data at each acquisition time. The LOF anomaly detection algorithm is a well-known technology and will not be described in detail in this application. As other implementation methods, based on the ability to measure the degree of anomaly of each parameter data at each acquisition time, implementers may use other existing technologies, such as the isolated forest algorithm, etc. This application does not impose any special restrictions.

[0032] Furthermore, to accurately analyze the comprehensive change characteristics of different parameters fluctuating over time, for the parameter feature matrix, the DTW (Dynamic Time Warping) distance of the parameter vector between each row of data and every other row of data in the parameter feature matrix is ​​calculated. The mean of the distances between each row of data and all other rows of data is calculated, and the normalized value of the mean is used as the difference weight of each row of data. The sum of the difference weights of all rows of data in the parameter feature matrix is ​​1. The DTW distance is used to measure the degree of difference between each row of data and every other row of data; the larger the DTW distance, the greater the degree of difference between each row of data and every other row of data.

[0033] In this embodiment, the Softmax function is used to obtain the normalized value of the mean. The Softmax function is a well-known technique and will not be described in detail here.

[0034] Furthermore, the difference weights of all rows of data in the parameter feature matrix are arranged in row order to form a weight matrix, which is a row vector. Using the weight matrix and the parameter feature matrix, the comprehensive characteristics of the abnormal changes of all parameters at each sampling time are analyzed to obtain the comprehensive feature matrix of the baffle distillation column. Specifically, the product of the weight matrix and the parameter feature matrix is ​​used as the comprehensive feature matrix of the baffle distillation column. The elements in the comprehensive feature matrix represent the comprehensive characteristics of the abnormal changes of all parameters of the baffle distillation column at each sampling time.

[0035] Step 2.2: Based on the mutation patterns of elements in the comprehensive feature matrix, the entire time period for collecting parameter data is divided into various time intervals.

[0036] Based on the above calculation results, the abnormal change characteristics of all parameters were analyzed in depth. Furthermore, the differences in abnormal change characteristics between different parameters were combined to evaluate the comprehensive abnormal characteristics of all parameters. If the parameters of the baffle distillation column show strong change characteristics for a long period of time due to interference, the comprehensive abnormal change characteristics at each sampling moment will increase significantly within the corresponding time range.

[0037] Based on the above analysis, all elements in the comprehensive feature matrix are arranged chronologically to form a feature sequence. A mutation point detection algorithm is used to obtain each mutation point in the feature sequence. The entire time period of the collected parameter data is divided into time intervals based on the time of each mutation point, where the time of the mutation point is the endpoint of the time interval. For example, the entire time period up to the current time is... The mutation points occur at times b and c, and The resulting time interval is: , and The purpose of determining the time interval segmentation point by utilizing the comprehensive characteristics of the abnormal changes of all parameters at different acquisition times is to fully consider the interval characteristics of abnormal changes of different parameters due to coupling relationships under the influence of interference. By using the relative differences in abnormal changes between different parameters, the significant abnormal change characteristics caused by interference and coupling relationships at each acquisition time can be determined more accurately. That is, if at a certain acquisition time, the interference causes abnormal changes with coupling relationships between different parameters, the calculated comprehensive characteristics of abnormal changes will be more significant. A schematic diagram of the time interval segmentation process is shown below. Figure 3 As shown.

[0038] Step 2.3: By comparing the values ​​of the comprehensive feature matrix within any time interval with the differences in the abnormality levels of each parameter data within the same time interval, the feature values ​​of each parameter data within that time interval are obtained; by the degree of overlap between the time interval and the time period, the feature factors of the time interval are obtained; and by combining the feature values, the feature coefficients of each parameter data within that time interval are obtained; by the distribution of the abnormality levels of each parameter data within each time interval and the feature coefficients, the control coefficients of each parameter data are obtained.

[0039] Based on the time interval segmentation results, the coupling characteristics of parameters in different parts of the baffled distillation column during the preparation of high-purity silane gas can be effectively compared when affected by disturbances, thereby determining the control coefficients of the parameters in each part. The specific process is as follows: The average of all abnormal scores for each parameter data within each time interval is calculated. A larger average value indicates more significant abnormal changes in the parameter data caused by interference and coupling correlation characteristics within each time interval. Further, to accurately analyze the significance of coupling changes caused by interference in each parameter data within each time interval, this application arranges all values ​​of the comprehensive feature matrix in any time interval in chronological order to form a first vector; arranges all abnormal scores for each parameter data within any time interval in chronological order to form a second vector for each parameter data; the DTW distance between the first vector and the second vector within any time interval is denoted as the metric distance; a larger metric distance indicates more significant abnormal changes in the coupling characteristics of each parameter under interference; the proportion of the metric distance corresponding to each parameter data within any time interval in all metric distances is used as the characteristic value of each parameter data within that time interval. A larger characteristic value indicates that, by comparing the abnormal changes in each parameter data with the overall characteristics, the abnormal changes in coupling characteristics caused by interference are more significant within that time interval. The DTW distance is used to measure the difference between the data in the first vector and the second vector. The larger the DTW distance, the greater the difference between the data in the first vector and the second vector.

[0040] The intersection-union ratio of any time interval with the time period is used as the characteristic factor of any time interval. The larger the characteristic factor, the greater the sustained influence of the parameter data in the time interval with significant abnormal changes. Considering the comprehensive characteristics of abnormal changes in coupling features under the influence of interference, which are both abnormally significant and persistent, the mean of the characteristic values ​​of each parameter data in any time interval and the characteristic factor of any time interval are used as the characteristic coefficients of each parameter data in any time interval.

[0041] Based on the time interval segmentation results, the control coefficients of each parameter data are obtained through the feature coefficients of each parameter data in each time interval, and the distribution of the abnormal scores of each parameter data in each time interval. The expression is as follows: In the formula, This represents the control coefficient for the x-th parameter; n represents the number of time intervals. This represents the characteristic coefficient of the x-th parameter data in the i-th time interval; This represents the average of all the anomaly scores for the x-th parameter within the i-th time interval.

[0042] It should be noted that the larger the value of the control coefficient, the more significant the coupling characteristics of the x-th parameter under actual control, and the greater the likelihood that the PID controller will not respond promptly to changes in the coupling characteristics under the influence of disturbances. A schematic diagram of the control coefficient acquisition process is shown below. Figure 4 As shown.

[0043] By comprehensively analyzing the abnormal changes in parameter data at different points during the control process of the diaphragm distillation column, the time intervals with interference and coupling characteristics are precisely divided. Then, the characteristics of untimely actual control response caused by strong coupling under the influence of interference are accurately analyzed, thereby achieving precise optimization and adjustment of the diaphragm distillation column.

[0044] Step 3: Using the control coefficients and a PID controller, control the various parameters of the baffle distillation column.

[0045] The initial values ​​of the proportional parameters of the PID controller are obtained through PID parameter tuning methods when controlling various parameters. Then, using these initial values ​​and the control coefficients for each parameter, the optimized values ​​of the proportional parameters are obtained. Specifically: The sum of the initial value and the normalized values ​​of the control coefficients of each parameter data is used as the optimized value of the proportional parameter of the PID controller when controlling each parameter.

[0046] In this embodiment, the PID parameter tuning method is the attenuation curve method. The attenuation curve method is a well-known technology and will not be described in detail here. As other implementation methods, based on the availability of the initial values ​​of the proportional parameters when the PID controller controls various parameters, the implementer may adopt other existing technologies, such as the trial and error method. This application does not impose any special restrictions.

[0047] In this embodiment, the Min-Max normalization method is used to obtain the normalized value of the control coefficient. The Min-Max normalization method is a well-known technique and will not be described in detail in this application.

[0048] After obtaining the optimized values ​​of the proportional parameters when the PID controller controls various parameters, the PID controller controls various parameters of the baffle distillation column based on the optimized values.

[0049] In summary, this application analyzes the degree of anomaly of each parameter at each acquisition time and constructs a parameter feature matrix to centrally reflect the anomaly characteristics of the parameter data. By calculating the difference weights of each row of data in the parameter feature matrix, the differences between different parameter data are quantified, providing a weighting basis for subsequent comprehensive analysis. Combining the difference weights with the parameter feature matrix, a comprehensive feature matrix is ​​obtained, which can comprehensively reflect the comprehensive characteristics of the anomaly changes of the baffle distillation column at different acquisition times. By applying the difference weights, the reasonable weights of each parameter data in the comprehensive analysis are ensured, improving the accuracy of the analysis of the comprehensive characteristics of the anomaly changes at each acquisition time. Furthermore, by comprehensively analyzing the abrupt changes of elements in the feature matrix, the time intervals of abnormal changes caused by interference and coupling relationships are precisely segmented. By comparing the abnormal changes of various parameter data within the segmented time intervals with the overall characteristics, the control coefficients of various parameter data of the baffle distillation column are determined. The control coefficients can reflect the interference influence and coupling characteristics of various parameter data in the actual control process, providing a basis for optimizing the control of the baffle distillation column. Through the control coefficients, the proportional parameters of the PID controller are adjusted. By accurately analyzing the characteristics of untimely actual control response caused by strong coupling relationships under the influence of interference, the timeliness of controlling various parameters of the baffle distillation column can be improved, the stability of various parameters of the baffle distillation column can be improved, and thus the quality of high-purity silane gas preparation can be improved.

Claims

1. A method for controlling a baffled distillation column for the preparation of high-purity silane gas, characterized in that, The method includes the following steps: Real-time acquisition of various parameter data of the baffle distillation column during the preparation of high-purity silane gas; By evaluating the degree of anomalies in various parameter data at each acquisition time, a parameter feature matrix of the baffled distillation column is constructed; by obtaining the difference weights of data in different rows of the parameter feature matrix, the difference weights of each row of data are obtained; and by combining the difference weights with the parameter feature matrix, a comprehensive feature matrix of the baffled distillation column is obtained. By analyzing the abrupt changes in the elements of the comprehensive feature matrix, the entire time period for collecting parameter data is divided into various time intervals. The feature values ​​of each parameter data within any given time interval are compared to the differences in the degree of abnormality of each parameter data within that time interval, thus obtaining the feature values ​​of each parameter data within that time interval. Furthermore, the feature factors of each time interval are obtained based on the degree of overlap between that time interval and the time period. Finally, the feature coefficients of each parameter data within that time interval are obtained by combining these feature values. By analyzing the distribution and characteristic coefficients of the anomalies of various parameter data within each time interval, control coefficients for each parameter data are obtained. By using the aforementioned control coefficients in conjunction with a PID controller, various parameters of the baffle distillation column are controlled. The difference weight is the normalized value of the mean DTW distance between each row of data and all other rows of data in the parameter vector, wherein the sum of the difference weights of all rows of data in the parameter feature matrix is ​​1; The process for determining the control coefficient is as follows: An anomaly detection algorithm is used to obtain the anomaly scores of each parameter data at each acquisition time, and the average value of all the anomaly scores of each parameter data in each time interval is calculated. Calculate the product of the characteristic coefficients of each parameter data in each time interval and the average value; The control coefficient is the sum of the products of the various parameter data over all time intervals.

2. The method for controlling a baffled distillation column for the preparation of high-purity silane gas as described in claim 1, characterized in that, The process of constructing the parameter feature matrix is ​​as follows: The anomaly scores of each parameter data at all collection times are arranged in chronological order to form parameter vectors for each parameter data, where the parameter vectors are row vectors; the parameter vectors of all parameter data are then combined to form a parameter vector matrix.

3. The method for controlling a baffled distillation column for the preparation of high-purity silane gas as described in claim 1, characterized in that, The process of obtaining the comprehensive feature matrix is ​​as follows: The difference weights of all rows of data in the parameter feature matrix are arranged in row order to form a weight matrix. The comprehensive feature matrix is ​​the product of the weight matrix and the parameter feature matrix, where the weight matrix is ​​a row vector.

4. The method for controlling a baffled distillation column for the preparation of high-purity silane gas as described in claim 1, characterized in that, The process of dividing the time interval is as follows: Arrange all elements in the comprehensive feature matrix in chronological order to form a feature sequence; Each mutation point in the feature sequence is obtained, and the time period is divided into time intervals based on the time of the mutation point.

5. The method for controlling a baffled distillation column for the preparation of high-purity silane gas as described in claim 2, characterized in that, The process for obtaining the feature values ​​is as follows: Arrange all values ​​of the comprehensive feature matrix within any time interval in chronological order to form a first vector; Arrange all the abnormal scores of each parameter data within any time interval according to time sequence to form a second vector of each parameter data; The DTW distance between the first vector and the second vector is denoted as the metric distance; The proportion of the metric distance corresponding to each parameter data in any given time interval among all the metric distances is used as the feature value of each parameter data in that given time interval.

6. The method for controlling a baffled distillation column for the preparation of high-purity silane gas as described in claim 1, characterized in that, The characteristic factor is the intersection-union ratio of any time interval to the time period.

7. The method for controlling a baffled distillation column for the preparation of high-purity silane gas as described in claim 1, characterized in that, The characteristic coefficient is the mean of the characteristic value and the characteristic factor.

8. The method for controlling a baffled distillation column for the preparation of high-purity silane gas as described in claim 1, characterized in that, The parameters of the control diaphragm distillation column include: The initial value of the proportional parameter of the PID controller when controlling various parameters is obtained by the PID parameter tuning method. The sum of the initial value and the normalized value of the control coefficient of each parameter data is used as the optimized value of the proportional parameter of the PID controller when controlling various parameters. The optimized values ​​are used to control various parameters of the baffle distillation column.

Images